Author: Yaspal Badyal, Rabten Datsang, Tianjun Ma, William Song 👨🔬
Affiliation: MVision AI, Virginia Commonwealth University 🌍
Purpose: Deep learning (DL)-based dose distribution predictions for prostate cancer show significant potential for OAR sparing compared to manually optimized treatment plans. We aim to generate clinically feasible, dosimetrically optimal plans guided by the predicted dose-volume parameters generated from a commercial DL solution.
Methods: Eight patient CT datasets, structure sets, and prescription dose information were input into MVision™ AI dose prediction solution. This prediction model is based on a modified CNN with U-Net architecture. The dose prediction results included 3D dose coverage and predicted OAR optimization objectives with gEUD and max dose parameters. These parameters, in addition to PTV objectives, were put into Eclipse v16.1 inverse optimizer to generate a clinically feasible plan. The resulting treatment plans were compared with manually optimized treatment plans in terms of dosimetric coverage, reduction in human interactions, and saved planning time.
Results: The target mean homogeneity index was 0.079 ± 0.021 and 0.087 ± 0.016, and the target mean conformity index was 0.954 ± 0.025 and 0.960 ± 0.005 for the manually optimized treatment plan and MVision™ AI dose prediction guided treatment plan (DL-guided plan), respectively. Average OAR mean doses were less for all DL-guided plans. Mean doses of Rectum, Bladder, and Penile Bulb were reduced by 9.64 ± 3.26Gy, 2.92 ± 2.37Gy, and 5.54 ± 5.66Gy, respectively. Total number of objective changes during optimization was reduced by 38.3 ± 24.5%. The planning time was reduced by 13.2 ± 14.5%.
Conclusion: DL-guided planning showed benefits in terms of time saved during the planning process and reduced human interaction with the optimizer. All the DL-guided plans had lower OAR mean doses and comparable target coverage, although the max doses were higher for some cases. Overall, the MVision™ AI dose prediction solution does help achieve better treatment plans, faster.